Enhanced Prediction of Immune Checkpoint Blockade Response in Melanoma Using Ensemble Learning and a Hybrid Feature Selection Technique

Immune checkpoint blockade (ICB) therapy provides significant clinical benefits to melanoma patients, however only a limited number respond effectively. The objective of this study is to develop a reliable and robust model to predict melanoma patients’ responses to ICB therapy. This is essential for guiding patient selection and optimizing treatment outcomes, ultimately supporting the goals of personalized medicine and improving recovery rates. The proposed model integrates hybrid feature selection, combining both filter and wrapper techniques, with an ensemble voting classifier to identify significant genes and enhance classification accuracy. We evaluated the performance of the model across five datasets of melanoma patients who received ICB treatment. The model successfully reduces feature dimensions to just 0.1% of the original set and achieves a prediction accuracy of over 89%. Notably, some of the gene signatures identified by our method have been previously reported in the literature as being associated with ICB response. Additionally, our model uncovered new significant genes, offering deeper insights into the underlying mechanisms of ICB therapy and helping to clarify the reasons behind the low patient response rate. Interestingly, our findings also reveal that the gene signatures extracted by our method are unique to each dataset, with no unified signature for ICB response prediction across different datasets. This aligns with previous assertions that global, generalized gene signatures for predicting ICB response do not exist, even within the same cancer type. Nevertheless, our approach offers a systematic and effective method for identifying the most informative genes, and provide high classification accuracy across datasets. The model demonstrates strong predictive capabilities and underscores the importance of personalized approaches, paving the way for further biological research.